CN1976629A - Medical Imaging System for Accurate Determination of Oriented Tumor Changes - Google Patents

Abstract

A body part (204) is scanned (20) to obtain a first set of image data (214A). A target lesion (5, 202A) in the image data is identified (30). The body part (204) is rescanned (40) at a subsequent time to obtain a second set of image data (214B). A target lesion (5A, 202B) in the second set of image data is identified, and the size of the target lesion (5, 202A) in the first and second sets of image data is measured to determine two apparent image volumes corresponding to the first and second sets of image data (60). The size change (70) is estimated by comparing the first and second apparent tumor sizes (301A, 301B). The variation in dimensional change is estimated (80) to determine the extent of the change in dimensional measurement.

Description

Medical Imaging System for Accurate Determination of Oriented Tumor Changes

technical field

The present invention generally relates to medical image data analysis, especially an automated computer method to accurately determine changes in target lesions or multiple lesions imaged by a medical imaging system.

Background technique

The pharmaceutical field develops a variety of products that require formal FDA approval, often based on measurements derived from medical imaging. The most costly and time-consuming aspect of drug development involves clinical trials to enable anticancer agents such as anticancer drugs to be approved. This is especially true for the field of oncology, but of course it also applies to other fields of medicine.

In the field of oncology, it is now commonplace to use medical imaging to assess response to treatment with anticancer agents. Many clinical trials use the measurement of changes in size of abnormal lesions or tumors (eg, tumors) as the primary indicator for assessing the efficacy of treatment. Although change in patient survival is considered the primary standpoint in assessing drug efficacy, this measure is (essentially) less important than change in tumor size, the measure that replaces it, as a means of obtaining FDA approval was given consideration. For example, a drug used to treat lung cancer might be rated based on the rate at which tumor or other growths in the lung decrease in size.

The RECIST (Response Evaluation Criteria in Solid Tumors) protocol is a formal, established method for measuring changes in tumor size. RECIST includes a set of published rules that define when a cancer patient's condition is considered to be improving ("responsive"), unchanged ("stable"), or severe ("deteriorating") during treatment. "). This standard was published through an international collaboration including the European Organization for Research and Treatment of Cancer (EORTC), the National Cancer Institute (NCI) and the Clinical Trials Group of the National Cancer Society of Canada. (See Therasse, et al., "New Guidelines to Evaluate the Response to Treatment in Solid Tumors," Journal of the National Cancer Institute, Vol. 92, No. 3, Feb. 2, 2000, 205-216.). Currently, most clinical trials assessing the objective response of solid tumors to cancer therapy use RECIST.

An element of the RECIST criteria is the use of single-space measurements in which the image containing the largest cross-sectional diameter of the tumor is selected and the one-dimensional largest measurement is derived from this image. This one-dimensional measurement is then compared to a similar image of the same tumor at another specific time to assess response. According to RECIST, a complete response was defined as disappearance of the tumor, a partial response was defined as a 30% reduction in size, and progression was defined as a 20% increase in tumor size. RECIST does not detect tumors smaller than 1 cm in size.

Accuracy is a critical issue when making any measurement. Unfortunately, the current RECIST method for assessing tumor response to treatment is very limited because it does not take into account the accuracy of the measurements. As a result, the method requires looking at a large number of changes in one-dimensional measurements to determine response to treatment. This need for such a large amount of variation makes the method unreliable for measuring tumor size.

Caliper measurements by radiologists can also be used to measure tumor size in previously standard practice. Measurement accuracy was assessed by skilled radiologists' measurement variability when measuring phantom or actual tumors. Errors associated with manual measurement of tumor length can be considerable. Similarly, since similar images cannot be reliably selected on instantaneously separated scans, large variations must be relied upon in order to be sure that the variation is real and not a measurement error.

In general, none of the current methods provide an accurate method with the steps of the present invention, which uses a volumetric method to determine size. Current methods measure tumor length in one slice and in only one or two directions, rather than measuring all three-dimensional elements associated with the tumor.

Contents of the invention

The present invention provides an automated method for determining the margin of error for volume change measurements. A certain part of the body is scanned to obtain the first set of image data. Identify orientation tumors in imaging data. The body part is then scanned again to obtain a second set of image data. A tumor in the second set of image data is identified, and a size of the tumor in the first and second sets of image data is measured to determine two apparent image volumes corresponding to the first and second sets of image data. Changes in size were estimated by comparing the size of the first and second sets of apparent tumors. Estimate the difference in dimensional variation to determine the range of variation in dimensional measurements.

In one aspect, the present invention provides a method of shortening the time of clinical trials by providing an accurate method of knowing whether a tumor is responding or not in a shorter time interval.

In another aspect, the present invention provides a method that can reliably measure changes in tumors to a lesser extent.

Description of drawings

Although the novel features of the present invention are described in detail in the claims, the following detailed description of the present invention (no matter for the structure or content) in conjunction with the drawings can be more helpful to understand and understand the present invention, and other purposes and objectives of the present invention features, among them:

Figure 1 shows a simplified block diagram of a system for the accurate determination and assessment of changes in tropisms imaged by an imaging system constructed in accordance with one embodiment of the present invention;

Figure 2 is a high level functionalized block diagram of an automated method for determining error bounds for imagery measurements constructed in accordance with one embodiment of the present invention;

FIG. 3 is a high level functionalized block diagram of another embodiment of a method for determining error bounds for image measurement constructed in accordance with an alternative embodiment of the present invention;

Figure 4 schematically shows CT image slices of the entire lung (large pulmonary);

Fig. 5 schematically shows the method of overlaying imaging on CT images to visualize tumor boundaries;

Figure 6 shows in schematic form an alternative method of visualizing tumor borders by superimposing imaging on CT images;

Figures 7A and 7B show in diagrammatic form another alternative embodiment of a method for visualizing tumor boundaries by superimposing imaging on CT images acquired at different times;

Figures 8A and 8B show in diagrammatic form yet another alternative embodiment of a method for visualizing tumor boundaries by overlaying images on CT images acquired at different times;

Figures 9A and 9B show in diagrammatic form another embodiment of a method for visualizing tumor boundaries by superimposing imaging on CT images acquired at different times;

Figure 10 shows in diagrammatic form another alternative embodiment of the method of overlaying imaging on a CT image to visualize tumor boundaries;

Figure 11 shows in diagrammatic form yet another alternative embodiment of the method of overlaying imaging on a CT image to visualize tumor boundaries.

Detailed ways

First, it should be noted that although the present invention describes in detail specific systems and methods for analyzing medical imaging data, such as radiology data, this is by no means limiting and is for illustration only, and the present invention can also be used to analyze other type data.

The present invention builds on the leading edge knowledge of imaging technologies that are now able to scan tumors so that the entire tumor volume can be imaged. Over the past decade, there have been major advances in methods for measuring tumor size from CT images by utilizing 3D volumetric computer algorithms. Furthermore, images are now acquired isotropically, ie the resolution is almost the same in the x, y and z directions. Advanced image processing allows for further segmentation of the tumor from surrounding structures, with better definition of tumor boundaries, resulting in improved measurements.

The present invention combines higher resolution imaging techniques with improved imaging methods to more accurately compare tumors. In this approach, changes to a smaller degree can be measured while still being confident in measuring changes in tumor size. Additionally, changes can be measured by measuring volume rather than simply using a single 1D measurement. This approach allows for a more complete evaluation of the data.

Measurement accuracy depends on many factors. By estimating the errors associated with these factors, the accuracy of any particular tumor measurement can be determined. As a result of knowing the accuracy of the measurements, a smaller range for changes in tumor size may be provided to predict severe disease. Therefore, severe disease can be diagnosed early and reliably by utilizing the accuracy analysis, using smaller but more accurate size changes of tumors, rather than using existing RECIST criteria.

An existing clinical trial data management system known as ELCAP Management System (EMS) can be advantageously used in the method of the present invention. EMS provided all aspects of the trial management system, including teleradiologist identification and computerized analysis of imaging data. EMS is characterized by more efficient and timely management of clinical trials, resulting in shorter trial times, more accurate data measurements and better quality control. The amount of data lost on clinical trial features due to lack of patient treatment monitoring where the trial is conducted can also be improved by using a web-based system with real-time feedback and reporting. In addition to automated methods, a semi-automated approach that requires radiologists to manually draw certain descriptive boundaries to define area or volume measurements would improve overall reproducibility and accuracy.

Referring now to FIG. 1, there is shown a simplified block diagram of an automated system for the accurate determination of changes in tropisms imaged by an imaging system constructed in accordance with one embodiment of the present invention. The video system 2 generates video data at different times _t1 and _t2 . The anomaly 5 in the image data generated at time t1 also appears in the subsequent image data generated at time t2 and is designated as anomaly 5A. Lesion 5 and lesion 5A are the same lesion, but it is considered here that lesion size is different at different times after treatment with anticancer drugs for exemplary purposes. The image data is processed by running image processing software 7 in the computer processor 6 . Target tumors can include tumors, nodules, and the like. The images may also include a correction device 10, which will be discussed in further detail below.

The medical imaging system 2 may advantageously comprise any known medical imaging system. Some useful known imaging systems include computed tomography scanners, magnetic resonance imaging, positron emission imaging systems, X-ray imaging systems for vascular interventions and angioscopy/angiography procedures, ultrasound imaging systems and equivalent medical imaging systems. The scanned target tumors 5 may advantageously include tumor types specified by the World Health Organization (WHO) and RECIST criteria, including breast, lung, melanoma, colon, ovary, and sarcoma.

In a useful embodiment of the invention, the software 7 operates automatically to accurately measure the size and volume of the tumor 5 . By this method, if there is a time difference between the acquisition of image data of the targeted tumor 5, the change in volume can be estimated. The method of the present invention determines the degree of error associated with each measurement in order to estimate the volume and ultimately the proportional change in volume. Automated methods performed under computer control provide precise reproducibility. The correction method estimates measurement errors caused by artifacts. Phantom, simulated, and real nodules are used to characterize the measurement accuracy of different nodules and their corresponding appearance in images such as CT images.

Various characteristics of tumors are evaluated to determine the measured change in a given volume based on its apparent volume. Measurements will vary based on the difference in tumor versus background signal. Measurement error variation may advantageously include estimates of various locations of a tumor (eg, a nodule with particular marginal characteristics).

Thus, a boundary definition given for a particular edge can be estimated from the measured variation. As described below, other factors may also be advantageously assessed, including the degree to which adjacent structures are connected to the tumour, and the possible impact of adjacent structures on the volume estimate. The characteristics of the measurement equipment can also include factors that affect the variation of the error. Volumetric measurements are also affected by the inherent resolution of the imaging system itself and the amount of noise present in the image.

spatial correction

Standard correction methods include phantom scans and measurements of the amount of noise, artifacts and image distortion. Models are man-made objects with known dimensions. These factors are spatially dependent due to the physical properties of imaging systems such as CT scanners. That is, the measurement error varies depending on where in the body the measurement is taken and the position of the body within the scanner. Current operations cannot take advantage of these factors, but a traditional conservative global distortion figure provided by the manufacturer can be used.

Through the study of the model, for a given imaging system, it is possible to advantageously establish a map that can characterize the degree of image distortion, artifacts and noise for all relevant parts of the human body. Once established, this map can be used to determine more accurate measurement error margins in tumor measurements.

error correction

Accurate computerized measurements of tumor size from CT images use an algorithm to determine the exact location of the tumor's junction with other tissue. The algorithm can handle many different types of tumors and use different strategies to solve different situations. The establishment of error estimation is based on the image form and the specific algorithm processing method adopted for this image. In one aspect of the invention, a database is generated for each identifiable image distortion, and a measurement error estimate is derived from the statistical variation within the database.

According to the present invention, the method for systematic error estimation includes (a) measurements on CT images of the calibrated phantom, and (b) measurements on multiple scans of the patient's actual tumor. In one useful embodiment, measurements of slowly growing tumors can be obtained by scanning at short intervals.

As another example, repeated images of a tumor can be scanned at very short intervals, regardless of growth rate, resulting in an error estimate based on a substantially constant tumor. Such duplicate images can be obtained during a biopsy, where multiple images of a tumor are obtained within seconds. Furthermore, when a human is involved in the measurement process, variation or error due to human factors can be captured through human observer trials involving phantoms or repeatedly scanned tumors.

Errors associated with a particular geometric location (eg, a nodule attached to the chest wall) can be estimated by multiple imaging a set of phantoms simulating that location. The variability between scans of the artificial model can be used to characterize the margin of error for each specific trial location.

A good characterization of the variability parameters of the scanner system can be obtained through multiple measurements on the phantom images. For example, scanner reconstruction properties such as the point spread function can be accurately determined through model studies and experimental analysis. However, the model data cannot model all locations because some nodules have subtle differences in density that are difficult to model. In this case, multiple scans of many such nodules without apparent growth can be used to create a nodule database. One way to do this is to compare two scans of the same tumor taken within a short time interval. The nodule database can then be used to measure the measurement variation between the two scans in order to estimate the measurement error of a given class of different nodules.

Certain imaging problems produce certain artifacts. For example, heart motion can cause fluctuations in the z-direction in the shape of the 3D image. Another example would be bones on top that would generate excessive noise. These and other special cases can be identified, and margins of error can be estimated from a database of similar cases.

Error Estimation When Human Intervention Is Required

In some difficult-to-image locations, radiologists may intervene in the nodule segmentation process. Further processing by computer algorithms then reconciles the differences between the radiologist's judgments between the two scans. Estimation of measurement variability owes special credit to the intervention of radiologists by building a database in this case. Once all sources of measurement error have been determined, the overall measurement error can be calculated.

Referring now to FIG. 2, there is shown a highly functional block diagram of an automated method for determining error margins for dimensional measurement variations, according to one embodiment of the present invention. An automated method for determining a margin of error for volumetric changes includes the following steps:

In step 20, scan the body part with an imaging system to obtain a first set of image data;

At step 30, at least one tumor is identified in the image data;

In step 40, re-scanning the body part to obtain a second set of image data;

At step 50, at least one tumor is identified in the second set of image data;

At step 60, at least one tumor imaged in the first set of image data and the second set of image data is measured to determine a first apparent tumor size corresponding to the first set of image data and a size corresponding to the second set of image data. Second apparent directional tumor size;

At step 70, the size change is estimated by comparing the first and second apparent tumor sizes; and at step 80, the variation in the size change is estimated to determine the extent of the change in the size measurement.

The step of estimating the variation in dimensional change in step 80 may advantageously include the results of an assessment of a number of factors affecting the accuracy of the measurement. Standard statistical methods can be used to estimate or determine the variability of the image measurements and other error measures discussed in this invention. Such techniques include, for example, linear regression, random effects models, and the like.

Factors affecting measurement accuracy include major sources of error such as nodule form, scanner parameters, patient factors, algorithms, and operator factors. Many of them are interactive. For example, the definition of nodule boundaries depends on the nodule tissue, the scanner's point spread function, patient movement, and other factors. Estimates of the variation in error are made using image models for error factors and deriving the parameters of these models from measurements on the image models and measurements of the patient, also by computer simulation. Paired observations of the same patient can also be used to reduce errors.

Examples of nodule shape factors include:

a. Density distribution characteristics, such as

i. uniform or variable distribution characteristics, and/or

ii. Features of solid or diffuse tissue.

b. Nodule geometry features such as

i. Spherical or complex shapes whose complexity can be estimated, e.g., as a ratio of surface area to volume, normalized to spherical (=1),

ii. the shape of the multiple components,

iii. cavity, and/or

iv. Small features similar to (close to) reconstruction resolution.

c. Whether surface features such as nodules are rough (ie, have a complex surface) or smooth, where a rough surface implies a high mean curvature.

Examples of scanner parameters include:

a. Reconstruction resolution further includes slice thickness, overlap and/or in-plane pixel (pixel) size,

b. X-ray energy (dose): kVp and mAs,

c. Reconstruction filter,

d. The speed of the gantry (gantry),

e. Workbench (pitch) speed,

f. a spatially varying point spread function, and/or

g. correction.

Examples of patient factors include the following:

a. The location of the scan area in the body,

b. Body size

c. Inhalation degree,

d. Inspiratory movement (especially at the bottom of the lungs),

e. Minor muscle spasms,

f. The top of the lung, eg, streaking artifacts, and/or

g. Health status of the lung tissue adjacent to the nodule, noting the presence of scarring, emphysema, or other health-related conditions.

The operator factor comes from the operator, who plays an auxiliary role in the nodule measurement process. For example, the operator may have to manually vary the estimated nodule extent, introducing a measurable contribution to measurement error, which can be characterized by observer studies.

Fully automated algorithms usually have close positions to intrinsic decision points. For example, an automated algorithm might treat a mass around a tumor as a connected blood vessel or as part of a nodule. Algorithms can be designed to indicate how close they operate to a decision point, thereby indicating the error factor associated with falling on the other side of the decision point.

Once an imaging region has been determined to represent a nodule, the measured variability can be estimated by factoring in the imaging model as described below:

1. Density: Low variability is associated with a uniform solid tissue density distribution. High variability is associated with high image noise and low or spatially varying density distribution.

2. Shape: Low variability is associated with a spherical shape and high variability is associated with a highly irregular shape containing many masses or cavities.

3. Surface features: At the border (margin) of nodules, low variation is associated with high image gradient, and high variation is associated with low image gradient. Lower variance is associated with smooth surfaces, while high variance is associated with irregular surfaces with high curvature. Border regions between nodules and other associated solid structures such as blood vessels or the chest wall (where there is little or no evidence of bordered image gradients) must be treated differently. For low variation, these boundary regions should match between the two scans from the image segmentation algorithm. Since these regions cannot be determined as accurately as gradient edges, the ratio of non-gradient to gradient edge surface areas is directly related to the variation. The assignment of boundaries and incorporation of boundary accuracy in accordance with the present invention are discussed herein with reference to FIGS. 4-11 below.

4. Size: In general, the larger the nodule, the smaller the fraction of voxels and the more accurate the volume estimation. Low variability is associated with large nodules (or very fine scanner resolution), whereas large variability is usually associated with smaller nodules (assuming similar structural complexity (shape)).

Situations where estimated variation may be used include:

A. When both scans are available, all imaging data and parameters are considered to provide a range of estimated growth rates.

B. When only one scan was available, the estimated variation was used to determine the minimum time to wait for a second scan in order to arrive at a clinically important decision. It is the time at which the rate of malignant growth is measured within the margin of measurement error.

In some cases, dimensions are measured on a single image rather than a two-dimensional (2D) area of a volume estimated from a set of images.

In a preferred embodiment of the invention, each step is performed via suitable software that allows for the participation of medical personnel. An embodiment of the invention further comprises the step of defining at least one oriented tumor margin in the image data. Edge definition can be determined by applying a threshold and/or a gradient function to at least one anomaly to determine the boundary of the edge. To further aid in diagnosis, the software used employs automated segmentation and grading techniques known in the art to determine boundary and segmental features, including abnormalities, from body parts (eg, lungs) imaged by the imaging system.

In another embodiment, the method of the present invention includes the step of automatically estimating the degree of motion of a particular structure. In another useful embodiment, the method of the invention includes the step of automatically estimating the degree of motion for a particular structure, including measuring the degree of change in surface structure and structure external to the tumor. In the lungs, the degree of motion varies significantly depending on the location of the target tumor and the distance from the heart.

In another useful embodiment, the method of the present invention includes the step of automatically matching corresponding images of at least one tumor obtained at different times. For example, the software might select the aneurysm having the largest size in the image and compare it to a comparable aneurysm in the second, subsequently acquired image. Dimensional measurements may advantageously include length, area and three-dimensional volume of the tumor.

In another useful embodiment, the method of the present invention comprises the step of selecting as a target at least one tumour, which has the largest area, and finding a comparable target obtained at a subsequent time.

In another useful embodiment, the method of the present invention includes the step of spatially correcting the imaging system using at least one model, noise measurements, scanner artifacts, and image distortions.

Referring now to FIG. 3 , there is shown a highly functional block diagram of a method for determining the margin of error for image measurement. According to one embodiment of the invention, the steps of the method include:

In step 120, scan the body part with an image volume system to generate a set of image data;

In step 130, measuring at least one directional tumor imaged in the set of image data, thereby determining the apparent directional tumor size corresponding to the set of image data;

At step 140, estimating at least one error change in the size of the first apparent directional tumor to determine the estimate of overall measurement accuracy;

At step 150, using said estimate of overall measurement accuracy to determine a range for anomaly size; and

At step 160, a time frame is determined based on the estimate of overall measurement accuracy for a second measurement, indicative of clinical change.

The methods involved in this aspect of the invention are preferably performed using software installed on a personal computer. In a preferred embodiment of the invention, the change in size of an orientation tumor indicative of severe disease is less than that defined by RECIST criteria. The step of estimating at least one error parameter advantageously comprises (a) calculating an image measurement error of the corrected model obtained by a computed tomography scanner, and (b) calculating a measurement error of multiple scans of the tumor of the patient.

In another useful embodiment of the invention, the software employed further includes a processing program module to obtain the variation due to human intervention using data from human experiments with phantoms or repeatedly scanned tumors of known size. Experiment as an observer.

In one embodiment, the set of error factors includes at least one factor selected from the group consisting of:

The point spread function of the imaging device and the associated reconstruction filter;

Scanner parameters;

Artifacts caused by dense objects in the same image plane as the nodule;

patient movement;

Changes in patient position between scans;

Changes in physical condition or inspiratory volume when the lungs are scanned;

the size of the nodule;

Confusing structures associated with nodules;

Changes in nodule density;

Scanner calibration;

definition of nodule boundaries; and

Operator variability introduced when skilled workers are involved in the measurement process.

The scanner point spread function can be estimated from a set of test scan results obtained using a calibration model. The scanner point spread function can also be scanned using a 3D calibration model of the patient. Since the model dimensions are known, the scan provides information for estimating any bias due to the scan operator parameters. This bias information can then be used in the image data to minimize errors due to scan operator parameters.

By using a set of phantom scans with two parameter settings, it is possible to measure different scan worker parameters between at least two scans, thereby estimating volumetric deviations due to the different parameters. The ideal operation is to use two scan results with the same parameters.

Artifacts can advantageously be characterized by calculating an image noise index based on the spatial frequency composition in a target site of interest, such as a nodule or a tumor. Artifacts can also be characterized by data from consistency studies with models that have similar noise indices and other parameters that provide estimates of variation.

Patient movement during the scan can affect the results. Common forms of patient motion include cardiac motion, patient muscle spasm, respiration, pulse oscillations, or other forms of patient motion during scanning of a tumor such as a nodule. For example, patient motion errors characterized by heart beats are detected by repetitive changes in the z-axis direction on the imaged nodule surface. In addition to patient motion, patient position also affects imaging results. The change in the patient's position between scans is measured by comparing the fit of the 3D rigid body's position between at least two scans taken at different times.

Changes in a patient's condition can be measured by 3D matching between the two scans. The variation in breathing error can be estimated by studying the scan pair data table. For large changes in respiration, a study of the scan pair data table can be used to estimate the bias and variability caused by this condition.

When the nodule is a nodule, nodule size error can often be characterized by phantom studies using different sizes to determine the inherent measurement error for a given nodule size. Similarly, errors due to connected structures can be characterized by model data from multiple scans and the measured variation of connected structures under different conditions. Connected structures include, for example, organ junctions or appendages of organs of similar density. Errors due to connected structures can advantageously be characterized by data from nodules of known size with multiple scans and attachments for comparison of segmentation consistency.

Errors due to scanner corrections can also be characterized by histogram adjustment using image noise in local image statistics. Errors due to scanner corrections can also be characterized by using a correction model scanned with body parts.

In one embodiment, errors due to nodule boundary definition can be characterized by comparing nodule boundary maps to point spread functions. Errors due to nodule boundary definition can also be characterized by performing modeling studies to determine variation in volume estimation under different conditions. Errors due to nodule density can be characterized by comparing multiple scans of slow-growing tumors of known size.

In another embodiment, error due to operator variation can be measured by a human observer study involving a number of radiologists and estimating these human variations under different imaging quality conditions.

As mentioned above, there are various sources of error when making measurements. Selecting a certain mode of operation when scanning, such as keeping the slice thickness constant, can control certain error factors. Other factors are intrinsic to the scanner, such as the modulation transfer function (MTF) of the scanning system. In some cases, such intrinsic factors as MTF may be specified by the scanning system manufacturer. Currently, there are no generally accepted standards for imaging methods for cancer-related measurements. However, the effect of the error factor on measurement accuracy is sufficient to raise a given level of confidence using the error variance measure, and the measurement accuracy measure may be estimated or derived as discussed herein. Another way to gain greater confidence in the accuracy of the measurements is to scan the patient each time with a calibration device.

Referring again to FIG. 1 , the present invention optionally includes the use of a calibration device 10 when volume estimation is to be performed whenever a patient is scanned. The correction device may comprise a phantom which is scanned at the same time as the patient is scanned. With this approach, the artificial phantom will be subjected to the same scan parameters as the patient. The correction device can be placed at the scan center, and/or, in addition, the correction device can be given to the patient to take it with them. The calibration device advantageously contains a set of artificial phantoms of different sizes. The man-made model may include a set of highly corrected spherical objects, and a set of more complex structures.

In one embodiment, the calibration device can be placed in an acrylic or plastic cast, which can be relatively small. For example, a device in the size range of about 2cm x 2cm x 2cm that is easily portable about the size of a standard envelope, an ordinary book or the like may be used depending on the desired size. Larger or smaller devices may also be appropriate in certain scanning situations. Other calibration devices may include wires, beads, rods, or similar items of known size and/or density. The device can be placed on the patient and subjected to the same scan parameters as it is being scanned. Objects within the model are then measured. Using multiple objects of different sizes and types (highly corrected for size and density) a measurement of variation can be performed to account for bias and repeatability. In this approach, the measurement accuracy of a given scanner can be estimated by using specific instrument settings when scanning a given patient. Measurement accuracy is further improved by using other known information about the scanning device, intrinsic factors such as MTF discussed above.

An alternative embodiment of the method of the present invention may use an in vivo calibration device or device. For example, wires, beads, catheters, implantable devices, or similar items of known dimensions can be placed inside a patient for correction or other medical purposes. Such in-vivo devices or elements may be used to correct scans and correlate scans obtained at different times with errors, between different scans, or both.

Referring now to FIG. 4 , this figure is a slice through a CT image of a large pulmonary nodule. The CT image 214 shows a pulmonary nodule 202 comprising substantially coherent masses in a region 208 of the lung 204 . Other physical features include the spinal cord portion 206 and other features 210 and 212 connected to the lungs. Lung nodules 202 typically include spicules emanating from the nodules, as is clearly shown in FIG. 6 .

Those skilled in the art appreciate that typical CT images often do not clearly show defined boundaries of tumors (eg, nodules) and surrounding features.

Referring now to FIG. 5 , a nodule boundary visualization method is generally shown on an overlay imaged CT image. In the preferred embodiment, the various dashed lines 218, 220, and 222 represent color-coded boundaries, illustrating regions of error origin. In one embodiment, dashed line 220 may correspond to a light green boundary, indicating a well-defined nodule edge (eg, with a height image gradient); thus, the expected error for the green boundary is small. Dashed line 222 corresponds to a light red border, indicating that this region has low image gradients, or that this region has very subtle features of a tumor (called spicules) and therefore can be ignored from volume estimation. The presence of low image gradients or needle loading reduces the accuracy of the measurement. The dashed line 218 corresponds to the light blue border, illustrating that this region has little or no evidence of image gradients for the border. In such cases, radiologists may be allowed to use interactive software to manually determine the position of the border. Regions with low image gradients are the source of the largest boundary localization errors. In this way, the placement accuracy of nodule borders is visualized, indicating the source and likely magnitude of errors.

In one embodiment of the invention, the color-coded boundaries can be automatically drawn on the display by known graphics software techniques in conjunction with the information disclosed in the present invention. For example, colors may be selected based on the error associated with a given boundary (determined by the edge-finding software) and the associated error variation or other parameters (determined in accordance with the above disclosure). Appropriate keywords or instructions can be given to help the operator interpret the displayed results or images.

Referring now to FIG. 6 , another alternative embodiment of a nodule boundary visualization method is shown schematically overlaid on a CT image. In this alternative embodiment, nodules may advantageously include colored borderlines including, for example, yellow 224, light blue 218, light green 220, and light red 222, where the colors are represented by different types of dashed lines. The double boundary lines surrounding the region 208 can be used to illustrate the estimated range of error. That is, true nodule boundaries were considered to be within the double boundary line. In this example, yellow 224 is used to delineate subtle features, such as spicules, which are considered constituents of nodules, but are ignored when the measurement process involves nodule volume, as they are also measurements large source of error.

Such needles include complexes (but not medically insignificant structures), which are statistically processed as outliers according to the method of the invention. Such structures are usually long and thin, but small in size. Typically, small volume structures associated with height errors are ignored so as not to distort measurement accuracy results.

Referring now to FIGS. 7A and 7B , another alternative embodiment of the nodule boundary visualization method is shown schematically overlaid on CT images acquired at different times. Here, a solution similar to the visualization solution mentioned above is used, wherein at least two scans of the tumor are available and wherein the differences between the scans are also visualized.

FIG. 7A shows a first CT image 214A of a nodule 202A taken for the first time, and FIG. 7B shows a second CT image 214B of the same nodule 202B taken a second time. Color-coded borders 218A, 220A, and 222A are applied to first image 214A, and then the techniques described above with reference to FIGS. 5 and 6 are employed. Color-coded borders 218B, 220B, and 222B are applied to second imagery 214B, and the techniques described above with reference to FIGS. 5 and 6 are then employed.

Referring now to FIGS. 8A and 8B , another alternative embodiment of a nodule boundary visualization method is shown schematically overlaid on CT images acquired at different times in which the nodule grows or otherwise changes in size. is visible. Here, the boundaries 218A, 220A, and 222A of the first image are superimposed on the boundaries 218B, 220B, and 222B obtained from the second image. The resulting overlay is displayed (on a computer monitor or other suitable display) to provide a visual change in nodule size and an indication of the accuracy of the measurement against the color barcoded boundaries.

Those skilled in the art who benefit from the disclosure of the present invention can understand that the boundary technology disclosed in the present invention is not limited to the embodiments. There are many possible variations of these visualization methods, including:

1. Label the 3D rendering of the nodule from all image slices,

2. Use of transparent (eg, colored) markers so that underlying structures can also be observed,

3. Mark with dark lines,

4. Use graduated markers so that distances can be observed quantitatively,

5. Incorporate distance scales and text descriptions to reflect quantitative measurements, and/or

6. Any combination of the above changes.

Referring now to FIGS. 9A and 9B , another alternative embodiment of a nodule boundary visualization method is shown schematically overlaid on CT images obtained at different times in which the nodule grows or otherwise changes in size. is visible. Here, cross-hatched regions 301A and 301B are regions where nodule size varies between the two CT images 214A and 214B. For example, the cross-hatched area may advantageously be displayed on a color monitor, for example in bright red. Other colors can also be applied.

Referring now to FIG. 10 , another alternative embodiment of a nodule boundary visualization method is shown schematically overlaid on a CT image representing another example in which nodule growth or other dimensional changes are visible. Here, cross-hatched areas 303, 305, 307, and 309 may be displayed in various colors, illustrating that a change has occurred, and with what degree of confidence. In one embodiment, region 303 corresponds to the yellow zone, which represents a higher degree of variation relative to the original tumor size estimate. Region 307 corresponds to the green zone, involving the uncertainty of the size associated with the second measurement. Area 307 corresponds to a red zone, representing a high probability of change in the zone. Area 309 corresponds to the blue zone, which represents the area where some measurements overlap. This diagram also sets up a model for what we do when we want to measure the response. Furthermore, the center point 320 can be advantageously chosen so that the variation estimates can be made for the various quadrants 320A, 320B, 320C and 320D of the block. In some volumes the changes may be much larger than others, and with different degrees of certainty.

Reference is now made to FIG. 11 , which is an alternative embodiment of a nodule boundary visualization method shown schematically overlaid on a CT image, also representing another example in which nodule growth or other dimensions Changes are visible. Figure 11 is essentially the same as Figure 10, with the addition of a boundary line 313 which is put together with that part of the nodule where change cannot be reliably measured (instead no change can be reliably determined). This enables other parts of the nodule to be analyzed.

While particular embodiments of the present invention have been illustrated and described, many modifications and changes in this invention will occur to those skilled in the art. It is, therefore, to be understood that the appended claims are intended to protect all such modifications and changes which fall within the subject matter and scope of the invention.

Claims (58)

1. An automated method for determining the error range of a dimensional change measurement, the method comprising the steps of: Use the imaging system (2) to scan (20) body parts to obtain the first set of image data; identifying (30) at least one orientation tumor (5) in the imaging data; Rescanning (40) the body part to obtain a second set of image data; identifying (50) at least one tumor in the second set of image data; measuring (60) at least one tumor (5) imaged in the first set of image data and the second set of image data to determine a first apparent tumor size corresponding to the first set of image data and a size corresponding to the second set of image data the second apparent directional tumor size of the data; estimating the change in size by comparing the first and second apparent tumor sizes (70); and Variation in the dimensional change is estimated (80) to determine a range of variation in the dimensional measurement. 2. The method of claim 1, wherein the dimensional measurement comprises at least one dimensional measurement selected from the group consisting of length, area or three-dimensional volume of the tumor. 3. The method according to claim 1, further comprising the step of defining the margin of at least one tropism (5) by adjusting the apparent image volume of the at least one tropism (5) according to the image measurement variation, so as to generate at least two Adjusted image volume. 4. The method according to claim 3, wherein the step of defining the edge of at least one burr (5) further comprises the step of applying a threshold and/or a gradient function to the at least one burr to determine the boundaries of the edge. 5. The method according to claim 1, wherein the step of estimating the imaging measurement variation (80) further comprises the step of measuring a number of features of at least one anomaly (5), measuring features of adjacent structures, imaging system features, so The imaging system characteristics described include its intrinsic resolution and the amount of noise present in the image. 6. The method according to claim 1, wherein each step is carried out by using suitable software (7) allowing the participation of medical practitioners. 7. The method of claim 1, further comprising the step of automatically estimating the degree of motion for a particular structure. 8. The method of claim 6, wherein the step of automatically estimating the degree of motion for a particular structure comprises measuring the degree of change in surface structures and structures oriented extratumorally. 9. The method of claim 1, further comprising the step of automatically matching corresponding images of the at least one tumor obtained at different times. 10. The method of claim 9, further comprising the step of selecting as a target at least one tumour, having the largest area, length or volume, and finding comparable targets obtained at subsequent times. 11. The method according to claim 1, further comprising the step of spatially correcting the imaging system with at least one model (10) and measuring the amount of noise, artifacts and image distortions. 12. The method of claim 11 , wherein the step of spatially correcting further comprises the step of performing a model study of relevant regions of the body for a given scanner to create a map characterizing the degree of noise, artifacts, and image distortion; and utilizing The graph determines the measurement error range for the measurement of the orientation tumor (5). 13. The method according to claim 1, wherein the body part is a lung (204), the method further comprising the step of automatically segmenting other features of the lung from at least one tumour. 14. The method of claim 1, wherein the imaging system is selected from the group consisting of computed tomography scanners, magnetic resonance imaging, positron emission imaging systems, x-ray imaging systems, vascular interventions and angioscopy/angiography procedures or ultrasound imaging system. 15. A method for determining the error range of image measurement, the method comprising the following steps: Scan the body part with an imaging system to generate a set of image data (120); measuring at least one tumor imaged in the set of image data to determine an apparent tumor size corresponding to the set of image data (130); estimating at least one error variation in the apparent directional tumor size to determine an estimate of overall measurement accuracy (140); Using this estimate of overall measurement accuracy to determine a range of targeted tumors (150); and A time frame based on the estimate of the overall measurement accuracy for taking subsequent measurements indicative of clinical changes is determined ( 160 ). 16. The method of claim 15, wherein the change in size of at least one tumour, indicative of a significant event, is smaller than specified by RECIST criteria. 17. The method of claim 15, wherein the step of estimating at least one error variation comprises (140) the steps of: (a) calculating a measurement error from a computed tomography image of the corrected phantom, and (b ) to calculate the measurement error from multiple scans of the patient's tumor. 18. The method of claim 15, further comprising the step of error variation due to human participation through human observer experiments using phantoms or repeatedly scanned tumors. 19. The method of claim 15, wherein the step of estimating at least one error variation (140) comprises estimating an error impact due to at least one factor selected from: The point spread function of the imaging device and the associated reconstruction filter; Scanner parameters; Artifacts caused by dense objects in the same image plane as the nodule; patient movement; Changes in patient position between scans; Changes in physical condition or inhalation when the lungs are scanned; the size of the nodule; Confusing structures associated with nodules; Changes in nodule density; Scanner calibration; definition of nodule boundaries; and Operator variability introduced when skilled workers are involved in the measurement process. 20. The method of claim 19, wherein the point spread function of the scanner is estimated from a set of test scans obtained using the calibration model. 21. The method according to claim 19, wherein the scanner point spread function is estimated from a scan using a 3D corrected model (10) of the patient. 22. The method of claim 19, wherein the different scan worker parameters between at least two scan results are measured using a set of model scan results using the two parameter settings, thereby estimating the Volume deviation. 23. The method of claim 19, wherein a scan of the three-dimensional model of the patient provides information to estimate bias, thereby reducing errors due to scanner parameters. 23. The method of claim 19, wherein the artifact is characterized by computing an image noise index based on the spatial frequency content of the nodule region. 24. The method of claim 19, wherein the artifacts are characterized by a consistency study using a model (10) with a similar noise index, and other parameters provide an estimate of the variation. 25. The method of claim 19, wherein patient motion errors are characterized by cardiac motion or pulse vibrations detected by repetitive changes in the z-axis direction on the surface of the imaged nodule. 26. The method of claim 19, wherein the patient motion is selected from heart motion or patient muscle spasm during nodule scanning. 27. The method of claim 19, wherein the change in patient position between the two scans is measured by comparing the matching of the 3D rigid body position between the at least two scans. 28. The method of claim 19, wherein changes in breathing error are estimated by using a study of a scan pair data table. 29. The method of claim 19, wherein nodule size errors are characterized by phantom studies using different sizes to determine intrinsic measurement variation for a given nodule size. 30. The method of claim 19, wherein errors due to connected structures are characterized by using model data of connected structures with multiple scans and measurement variations under different conditions. 31. The method of claim 19, wherein errors due to connected structures are characterized by real data of nodules with attachments and multiple scans for comparison of segmentation consistency. 32. The method of claim 19, wherein error due to nodule density is characterized by comparing multiple scans. 33. The method of claim 19, wherein errors due to scanner calibration are characterized by using two-study histogram matching. 34. The method of claim 19, wherein errors due to scanner calibration are characterized by using a calibration model at the time of scanning. 35. The method of claim 19, wherein errors due to nodule boundaries are characterized by comparing a nodule boundary map to a point spread function. 36. The method of claim 19, wherein error due to nodule boundary definition is characterized by performing a modeling study with the model to determine variation in volume estimation under different conditions. 37. The method of claim 1, wherein the step of scanning the body part further comprises scanning a correction device (10). 38. The method of claim 19, wherein the step of scanning the body part further comprises scanning a correction device (10). 39. An automated method for determining a margin of error in a dimensional change measurement, the method comprising the steps of: Scan the body part (204) with the imaging system (2) to obtain the first set of image data (214A); Identifying a majority of orientation tumors (5) in the imaging data (214A); re-scanning the body part (204) to obtain a second set of image data (214B); identifying a majority of orientation tumors in the second set of image data C; Measuring a majority of the orientation tumors (5) imaged in the first set of image data (214A) and the second set of image data (214B) to determine a first apparent orientation tumor score corresponding to the first set of image data and corresponding to a second apparent orientation tumor score of the second set of imaging data; Estimate the change in directional tumor score by comparing the first and second apparent tumor scores (70); and The variation in the change in the directional tumor score is estimated to determine the range of change in the score measure (80). 40. The method according to claim 15, wherein the at least one tumor (5) is smaller than 1 cm. 41. A method for visualizing nodule boundaries on a CT image with overlapping imaging, the method comprising the steps of: obtaining a first image (214A) of the nodule (208A); determining a first set of boundaries (218A, 220A, 222A) of nodules corresponding to image gradient levels; and displaying the first image while overlaying the boundaries (218A, 220A, 222A) on the first image (214A). 42. The method of claim 41, further comprising the step of using different colors for the boundaries (218, 220, 222, 224) to indicate boundary error value ranges. 43. The method of claim 41, further comprising the step of excluding needle information (224) from measurement accuracy calculations. 44. The method of claim 41, further comprising the step of using interactive software to manually determine the location of the selected boundary. 45. The method of claim 41, further comprising the steps of: obtaining a second image (214B) of the nodule (208B); applying the second set of boundaries (218B, 220B, 222B) to the second image (214B); and The first set of boundaries (218A, 220A, 222A) and the second set of boundaries (218B, 220B, 222B) are overlaid on the display to show any changes in nodule size. 46. The method of claim 45, wherein regions showing changes in nodule size are visually displayed. 47. The method of claim 41, wherein the first set of boundaries is displayed using a visualization method selected from the group consisting of: Label the 3D rendering of nodules from all image slices, Use transparent markers so that the underlying structure can also be seen, Mark with a dark line, Use graduated markers so that distances can be observed quantitatively, Incorporate distance scales and text descriptions to visualize quantitative measurements, and any combination of the above methods. 48. The method of claim 41, wherein areas on the image are overlaid with markers to illustrate selected image factors (303, 305, 307 and 309). 49. The method according to claim 48, wherein the selected imaging factors (303, 305, 307 and 309) are selected from regions of higher degree of variation associated with the initial tumor size estimate, Areas of uncertainty in size, areas of high potential for variation, and areas where overlap occurs in some measurements. 50. The method according to claim 48, the method further comprising the step of selecting a center point 320, thereby allowing estimation of changes in various quadrants (320A, 320B, 320C and 320D) of the nodule. 51. The method of claim 48, further comprising the step of placing boundaries (313) with the portion of the nodule where changes cannot be reliably measured, other than no changes can be reliably determined. 52. The method according to claim 1, further comprising the step of using a calibration device (10) during scanning. 53. The method of claim 52, wherein the calibration device comprises a man-made phantom. 54. The method according to claim 52, wherein the calibration device (10) comprises a set of man-made models of various sizes. 55. The method according to claim 52, wherein the correction device (10) is selected from wires, beads, rods or geometrically shaped objects. 56. The method according to claim 52, wherein the calibration device (10) is an in vivo device. 57. The method of claim 56, wherein the internal device is selected from wires, beads, catheters, implantable devices, or items within the patient's body of known size.

Images